Comparative Analysis of Deep Learning and Statistical Algorithms for Forecasting Inbound Foreign Tourism in Thailand

Accurate prediction of tourist arrivals and revenue is crucial for establishing effective tourism development strategies in Thailand. This paper aims to predict the number of foreign tourists visiting Thailand and the revenue generated from them using the Long Short-Term Memory (LSTM) algorithm in comparison with the popular statistical method in economics, the Autoregressive Integrated Moving Average (ARIMA). We utilized real regional and country tourist data sets, collected by the Ministry of Tourism and Sports, and performance metrics, normalized root mean square error (NRMSE) and normalized mean absolute error (NMAE). Results reveal that LSTM yielded NRMSE values better than ARIMA ranging from 1.18 times to 19.33 times, and NMAE values better than ARIMA ranging from 1.40 times to 21.70 times across all data sets, except for the revenue forecast using the central region data set where ARIMA outperformed LSTM. This suggests that economists should consider adopting LSTM alongside traditional statistical techniques.